Drop a side-on clip of yourself pedaling. AI tracks your joints, measures the angles a fitter checks, and tells you the exact change. Your video never leaves your device.
Side-on, 20–30s, phone at hip height, pedaling steadily.
Angles at the bottom of your stroke. Every target links to where it came from — some are from papers, some are only a fitter's rule of thumb.
Here's what a finished analysis looks like. Your own numbers replace this once you add a clip on the left.
Knee 30–40° & elbow 15–30° — joint-angle averages (knee 36±7°, elbow 19±8°) from the Holmes method: Anthropometrics, flexibility and training history as determinants for bicycle configuration, Sports Medicine and Health Science.
Why 30–40° and not a static 25–35° — angles measured from video while pedaling run about 8° higher than a static goniometer: Validity and reliability of different kinematics methods used for bike fitting.
Front / rear thresholds (beta) — healthy pelvis coronal ROM 7.1±2.5° and knee coronal ROM 6.6±2.7°: Cycling kinematics in healthy adults, BMC Musculoskeletal Disorders. Knee valgus by alignment group: Effects of knee alignments and toe clip on frontal-plane knee biomechanics in cycling, JSSM. The precision floor of 2D video (SEM 2.7–3.0°, smallest detectable difference 7.5–8.9°) is why no frontal flag sits below ~9°: Reliability of 2D video assessment of frontal-plane dynamic knee valgus.
Torso, shoulder and the per-bike targets have no paper behind them. They come from a practitioner reference (bikefitadviser) and are a rule of thumb, not research — treat them as softer than the knee number. Full list of sources.
Point the camera square to the view you picked — dead side-on, front-on, or straight behind. Off-axis skews the angles.
Phone at hip/crank height on something stable — nothing blocking your legs (or arms, side-on).
20–30s at an easy, normal cadence — a trainer is ideal. Good light = less blur.